Fees Paid to Audit Firms, Accrual Choices and Corporate Governance*
David F. Larcker Ph 215 898 5424
Email: [email protected]
Scott A. Richardson Ph: 215 898 2063
Email: [email protected]
The Wharton School University of Pennsylvania
Philadelphia, PA 19104-6365
Revised: October 28, 2003
*We appreciate the comments of Stanley Baiman, Jan Barton, Sudipta Basu, George Benston, Jeff Coulton, Richard Leftwich, Linda Myers, Grace Pownall, Stephen Taylor, an anonymous reviewer and seminar participants at Emory University. The financial support of The Wharton School and Ernst & Young LLP is gratefully acknowledged.
1
Fees Paid to Audit Firms, Accrual Choices and Corporate Governance
Abstract
We examine the relation between the fees paid to auditors for audit and non-audit services and the choice of accrual measures for a large sample of firms. Similar to Frankel et al. (2002), we find that the ratio of non-audit fees to total fees has a positive relation with the absolute value of accruals. However, using latent class mixture models to identify clusters of firms with a homogenous regression structure reveals that this positive association only occurs for about 8.5 percent of the sample. In contrast to this result, we find consistent evidence of a negative relation between the level of fees paid to auditors and accruals (i.e., higher fees are associated with smaller accruals). The latent class analysis also indicates that this negative relation is strongest for client firms with weak governance. Overall, our results are most consistent with auditor behavior being constrained by the reputation effects associated with allowing clients to engage in unusual accrual choices.
2
Fees Paid to Audit Firms, Accrual Choices and Corporate Governance
1. Introduction
The purpose of this paper is to examine the relation between the fees paid to audit
firms for audit and non-audit services and the behavior of accounting accruals. This
relation (if it exists) is an important input into the ongoing debate in the regulatory and
academic communities about the structure of the accounting profession and the
appropriateness of providing non-audit services by accounting firms. Critics contend that
the extensive fees paid to auditors, especially fees for non-audit services, increase the
financial reliance of the auditor on the client (e.g., Becker et al., 1998 and Magee and
Tseng, 1990). As a result, independence may be compromised because the auditor
becomes reluctant to raise issues with the preparation of the financial statements at the
risk of foregoing lucrative fees. In contrast, DeAngelo (1981), Simunic (1984), and
others argue that the auditor faces substantial economic costs when audit failures are
observed. Thus, the relation between audit fees and auditor behavior is theoretically
ambiguous.
Prior research has examined many facets of this research question, but there is
little evidence that the level of audit fees or the provision of non-audit services is
associated with earnings quality. Frankel et al. (2002) claim that there is a positive
relation between the provision of non-audit services and accrual measures, which implies
that non-audit services impair earnings quality. However, more recent work by Antle et
al. (2002), Ashbaugh et al. (2003), Kinney, Palmrose and Scholz (2003) and Chung and
Kallapur (2003) have cast serious doubt on the findings and interpretations in Frankel et
al. (2002).
3
There are at least three economic or econometric explanations for conflicting
results in the prior literature. First, the role of corporate governance has been largely
ignored in the research to date. The auditor is only one of many potential monitoring
mechanisms designed to mitigate the inherent agency problems in a publicly traded firm.
Examining the auditor in isolation of alternate governance mechanisms provides an
incomplete analysis of the determinants of earnings quality. Second, there are many
ways to measure the financial connection between the auditor and client. Prior research,
such as Frankel et al. (2002), has tended to focus on the provision of non-audit services
(e.g., ratio of non-audit fees to total fees). However, the total fees paid to the auditor are
an equally plausible measure for the dependence of the auditor on the client (DeAngelo,
1981, Reynolds and Francis, 2001, and Chung and Kallapur, 2003). Finally, different
models are likely that describe the relation between audit fees and earnings quality across
a large sample of firms. For example, the importance of the monitoring role served by
the auditor should vary depending on the strength of the client’s governance structure.
Hence, using a single (pooled) regression model across a sample that is composed of
different models is unlikely to provide an adequate assessment for the relation between
audit fees and accrual choices.
We address these limitations in prior research by using latent class mixture
models to examine the relation between several audit fee measures (our proxies for
auditor independence) and accrual measures (our proxies for earnings quality) for a large
sample of firms for fiscal years 2000 and 2001. We find that a positive association
between audit fees and unexpected accruals occurs only when audit fees are measured
using the ratio of audit fees to total fees paid to the auditor and unexpected accruals are
transformed using the absolute value (i.e., a non-directional measure). Moreover, the
4
statistically positive association between non-audit fees and accrual behavior only occurs
for approximately 8.5 percent of the total sample. This small cluster of firms, relative to
the remaining clusters, has a smaller market capitalization, lower book-to-market ratio,
lower institutional holdings, and higher insider holdings. Thus, concurrent weakness in
corporate governance appears to be an important determinant of the relation between
auditor independence and earnings quality.
Although these results are provocative, we find that the relation between auditor
independence and earnings quality is highly sensitive to the specific measures used in the
analysis. Rather than simply using the ratio of non-audit fees to total fees, we consider
four alternate measures of auditor independence. Specifically, we use the ratio of dollar
(both audit and total) fees paid by the client to the auditor scaled by total fee revenue by
the auditor and abnormal audit (and total) fees based on the expectation models
developed by Simunic (1984) and Craswell et al. (1995).
In contrast to our initial results and those reported in Frankel et al. (2002), we find
a statistically negative relation between auditor independence (using the four alternate
measures described above) and earnings quality. Moreover, the cluster of firms with the
most pronounced negative association is characterized by low market capitalization, high
growth prospects, less independent boards, low institutional holdings, and high insider
holdings. For these firms the auditor appears to be playing a key role in the governance
process to limit abnormal accrual choices. Collectively, our results suggest that auditors
are less likely to allow abnormal accrual choices for firms where they have the greatest
financial interest. Overall, our results are most consistent with reputation concerns being
the primary determinant of auditor behavior with respect to limiting unusual accounting
5
choices of client firms (similar to the discussion in Reynolds and Francis, 2001 and
Chung and Kallapur, 2003).
The remainder of the paper is divided into four sections. Section 2 provides a
review of the prior research examining the relation of payments to auditors and
accounting choices. Section 3 describes our sample, provides descriptive statistics and
justifies the use of latent class mixture models. The results are presented in Section 4 and
interpretations and conclusions are provided in the Section 5.
2. Prior Research
Early research examined the relation between audit fees and non-audit fees in an
attempt to identify the economies of scale that existed from the joint provision of these
services (e.g., Simunic, 1984 and Palmrose, 1986). Recent research, however, has shifted
focus onto the potentially detrimental aspects of the provision of non-audit services.
Frankel et al. (2002) find that the provision of non-audit services is associated with (i) the
likelihood of reporting earnings that meet or slightly exceed analyst expectations and (ii)
the magnitude of the absolute value of abnormal accruals. Frankel et al. (2002) interpret
these results as strong evidence that the provision of non-audit services reduces auditor
independence and lower quality financial information.
Subsequent research, however, has questioned the appropriateness of the
conclusions in Frankel et al. (2002). Ashbaugh et al. (2003) find that after controlling for
firm performance there is no longer a positive relation between the provision of non-audit
services and measures of unexpected or abnormal accruals for a sample of 3,170 firms.
Similarly, for a sample of 1,871 firms, Chung and Kallapur (2003) also fail to find any
evidence of a relation between measures of unexpected accruals and measures of auditor
6
independence (using a measure of client dependence as opposed to simply examining the
provision of non-audit services). In addition, Ashbaugh et al. (2003) find no statistically
significant association between firms meeting analyst forecasts and auditor fees, and
Francis and Ke (2003) find that the association between firms meeting analyst forecasts
and auditor fees is very sensitive to the choice of comparison group.
Most prior research estimates the relation between the provision of non-audit
services and accruals using a relatively simple regression model between these two
variables. One notable exception, however, is contained in Antle et al. (2002) who
examine the relations among audit fees, non-audit fees and abnormal (or unexpected)
accruals in a simultaneous equations framework. Utilizing 2,443 firm-year observations
from the United Kingdom for the period 1992-2000, they find that the relation between
abnormal accruals and non-audit fees is negative after simultaneously estimating the
determinants of audit and non-audit services and accruals. They also find similar results
using a restricted sample of 1,430 U.S. firms for the year 2000.
Other measures have also been used to examine the impact of the relationship
between the client firm and auditor on earnings quality. DeFond et al. (2002) using a
sample of 944 financially distressed firms for the year 2000 find no evidence of an
association between the issuance of a qualified audit opinion and the provision of non-
audit services. Ruddock et al. (2003) using a sample of 4,708 Australian firm-year
observations from 1993-2000 find no association between measures of accounting
conservatism and the provision of non-audit services. Their argument is that if the
provision of non-audit services encourages income increasing earnings management this
will manifest via a reduction in observed accounting conservatism.
7
Recent work by Kinney et al. (2003) examines the relation between earnings
restatements and the provision of non-audit services. Using proprietary data for auditor
fees for a matched sample of 174 firms for the period 1995-2000, they find no consistent
evidence of a positive association between audit firm fees for non-audit services and
restatements. Rather they find a negative association between the provision of tax
services and restatements. Observing an earnings restatement, however, is open to
several interpretations. The traditional view is that the earnings restatement is indicative
of severe earnings management by the firm and carelessness by the auditors.
Alternatively, it could be the result of an effective auditor imposing their will on the firm
and forcing the restatement.
A related stream of research examines the financial dependency of the auditor-
client relationship. DeAngelo (1981) develops a model where the auditor faces a conflict
of interest. The auditor has to choose whether to compromise independence in return for
retaining quasi-rents from key clients. The incentives to compromise independence
depend on client importance (typically measured as the ratio of fee revenue from a
particular client deflated by total fee revenue for the audit firm). One outcome of the
financial dependency is that auditors may sacrifice their independence for more important
clients. Reynolds and Francis (2001) test this prediction using client size as a proxy for
audit fees and find no evidence that economic dependence impacts the audit outcome.
Specifically, for a sample of 6,747 firms in 1996 they find that abnormal accruals are
lower when client dependence is greater and there is a marginally significant greater
likelihood of receiving a qualified audit opinion. These results suggest that litigation and
reputation risk prompt auditors to curb aggressive reporting practices of firms.
8
Finally, accrual behavior and corporate governance has been examined in several
recent papers. Klein (2002) documents for a sample of 692 U.S. firm-years covering
1992-93 that the presence of independent outside directors on the board and audit
committee is associated with lower levels of unexpected or abnormal accruals (in
absolute terms). Other examples of this type of research include Xie et al. (2002) and
Jenkins (2002) who examine, among other things, the relation between board and audit
committee composition and measures of unexpected or abnormal accruals. Similar to
Klein (2002), these papers find that the presence of outside directors on the board and
audit committee is associated with lower levels of unexpected or abnormal accruals (in
absolute terms). This research highlights the importance of incorporating corporate
governance in the research design. In particular, corporate governance has an impact on
the demand for auditing quality and payment of audit and non-audit fees to the auditor,
and can have an important impact on financial reporting quality.
In summary, the literature examining the relation between audit fees and/or non-
audit services with accrual behavior finds virtually no statistical evidence for a relation
between auditor independence and earnings quality. Moreover, the results and
interpretations in prior research are statistically fragile and quite sensitive to changes in
research design and variable measurement. We extend the existing literature by
examining a more complete set of measures for both earnings quality and auditor
independence, relaxing the assumption that a single regression model describes the
relation between auditor independence and earnings quality, and explicitly analyzing the
role of corporate governance on the auditor-client relation.
9
3. Methodological Approach
3.1 Sample
Our initial sample consists of 5,815 firm-years that report data on audit and non-
audit fees paid for fiscal years 2000 and 2001. These data were obtained directly from
Standard & Poors.1 In order to reduce the impact of very small firms and auditors, we
restrict our sample to clients of Arthur Andersen, Ernst & Young, Deloitte & Touche,
KPMG, and PricewaterhouseCoopers (i.e., the “Big Five”) plus Grant Thornton and BDO
Seidman (reducing the sample by 355 observations). Firms are retained for subsequent
analysis if they have sufficient Compustat data for computing the accrual measures used
in our analysis. We also exclude financial institutions from the sample (SIC codes
between 6000 and 6999). This reduces the sample by 357 observations. The 5,103 firm-
years (3,424 firms) in our final sample span many sectors of the economy (Table 1, Panel
A). The industries most represented in our sample are business services (16.46 percent of
the sample), chemicals (9.21 percent), electrical (9.21 percent) and industrials (6.75
percent). These percentages are similar to the breakdown for the Compustat population.
The mean (median) of operating cash flow is equal to four (seven) percent of
assets, for book-to-market ratio is equal to 0.78 (0.52) and for market capitalization of
about $2,806 ($271) million. These numbers compare with a mean (median) book-to-
market ratio of 0.92 (0.58) and a mean (median) market capitalization of $2,910 ($139)
million for all firms on Compustat for the 2000-2001 period.
3.2 Measurement of Auditor Independence
1 It is important to note that the SEC disclosures with respect to audit fees are limited. Total fees and audit specific fees are disclosed but the remaining “other fee” category is not well defined.
10
We employ five different measures of the fees paid to auditors. First, we
calculate the fee ratio as the ratio of non-audit fees to total fees paid to the auditor (the
sum of audit and non-audit fees). Similar to prior research (e.g., Frankel et al., 2002), the
mean (median) firm pays roughly the same amount for non-audit and audit services (see
Table 2). However, there is considerable cross-sectional variation in this measure
(denoted as RATIO). Although RATIO has some intuitive appeal for measuring the
financial linkage between an auditor and a client, the size of payments to the auditor is
not captured by this measure. That is, a client with one dollar of audit and non-audit
payments produces the same score as a client with ten million dollars of audit and non-
audit payments.
In an effort to incorporate payment size into our analysis, we focus on the
importance of a particular client to the audit firm. In particular, we measure client
importance as the ratio of fees paid by the client firm to the total revenue of the auditor
for that year. As in Chung and Kallapur (2003), we obtain the total fee revenue for each
auditor from Accounting Today. We calculate two basic measures of client importance.
NONAUDFEE, uses non-audit fees to compute client importance, and TOTFEE uses total
fees (both audit and non-audit). We use both measures because the quasi rents could be
greater for non-audit services than audit services. The mean (median) firm pays total fees
that are approximately three (one) percent of total revenue for their auditor and non-audit
fees that are approximately two (one-half) percent of total revenue for their auditor
(Table 2, Panel A). As would be expected, these three measures of payments to the
auditor exhibit large, positive correlations (Table 2, Panel B). In section 4.3 we introduce
our final two measures of auditor independence that focus on abnormal fee levels.
11
3.3 Measurement of Accruals
Measuring earnings quality is a difficult and controversial task because there are
divergent views as to what constitutes “high quality” earnings. We follow prior research
and examine accrual measures as a proxy for issues of earnings quality (e.g., Frankel et
al., 2002, and Myers et al., 2003). The accrual component of earnings contains estimates
and forecasts, and is therefore easier to manipulate than cash flows. Thus, the flexibility
offered via accruals makes it a useful measure for examining the quality of financial
reports.
Rather than simply examining total accruals, we are interested in identifying the
“unexpected” component of total accruals. A large research has attempted to identify the
“unexpected” (also called discretionary or abnormal) accrual component. Jones (1991) is
the standard technique used for this decomposition. Total accruals are regressed on
variables that are expected to vary with “normal” accruals. These models have been
estimated using a time-series approach for each firm (e.g., Jones, 1991), or they have
been estimated in the cross-section for each industry (e.g., DeFond and Subramanyam,
1998). Both approaches have their limitations. The time series approach assumes
temporal stationarity of parameter estimates whereas the cross-sectional approach
assumes homogeneity across firms in the same industry. Consistent with the claims in
Bartov, Gul and Tsui (2001) that the cross-sectional models are better specified for a
sample of audit qualifications, we adopt a cross-sectional model. Furthermore, the cross-
sectional model maximizes our sample size.
Attempts to decompose total accruals into expected and unexpected components
can always be criticized for misclassifying expected accruals as unexpected because the
12
model of expected accruals is incomplete (e.g., Bernard and Skinner, 1996). To address
this issue, we use a more advanced model that attempts to mitigate the misclassification
issue. The advanced model is similar to that employed in Dechow et al. (2003). Dechow
et al. show that this model (i) has far greater explanatory power than the cross-sectional
modified Jones model, (ii) identifies unexpected accruals that are less persistent than
other components of earnings, (iii) identifies unexpected accruals that detect earnings
manipulation identified in SEC enforcement actions, and (iv) identifies unexpected
accruals that are associated with lower future earnings and lower future stock returns.
However, we acknowledge that attempts to decompose total accruals are still subject to
the limitation of model mis-specification.
Our accrual model builds on the cross sectional modified Jones model discussed
in Defond and Subramanyam (1998). This model assumes that the change in revenues
less the change in accounts receivable is free from managerial discretion (i.e., credit sales
are assumed to be abnormal) and that capital intensity drive normal accruals. We include
two additional independent variables that have been shown to be correlated with
measures of unexpected accruals. First, we include the book-to-market ratio (BM). BM
is measure as the ratio of the book value of common equity (Compustat item 60) to the
market value of common equity (Compustat item 25 x item 199). BM is included as a
proxy for expected growth in the firm’s operations. We expect to see large accruals for
growing firms (see also McNichols 2000, 2002). Investment in inventory and other
assets are likely to accompany growth phases of the firm’s life cycle. Observing an
increase in inventory in this circumstance is not necessarily due to opportunistic
managerial behavior. However, the modified Jones model classifies such increases as
“unexpected”. Second, we include a measure of current operating performance.
13
Previous research has shown that measures of unexpected accruals are more likely to be
mis-specified for firms with extreme levels of performance (Dechow et al., 1995). We
therefore include current operating cash flows, CFO (Compustat item 308), as an
additional independent variable. The advanced model is estimated as follows:
TA = α + β1(∆Sales-∆REC) + β2PPE + β3BM + β4CFO + ε (1)
Total Accruals (TA) is the difference between operating cash flows (Compustat
item 308) and income before extraordinary items (item 123) as reported on the statement
of cash flows. ∆Sales is the change in sales (item 12) from the previous year to the
current year, ∆REC is the difference in accounts receivable (item 302) from the start to
the end of the year, and PPE is the end of year gross property, plant and equipment (item
7). All variables are scaled by the average of total assets using assets from the start and
end of the fiscal year (item 6). The residual value from this model is labeled Accruals,
the estimate of unexpected or abnormal accruals from our extended Jones model.
Inclusion of both BM and CFO is not without issue. It is likely that incentives to manage
earnings vary in response to growth opportunities and current operating performance.
Specifically, market expectations of future growth can place greater pressure on
management to engage in earnings management (Dechow and Skinner, 2000). In
addition, current performance can create incentives to engage in earnings management.
Including these additional variables may be controlling for some of the variation in total
accruals that we are seeking to identify.2
2 It is appropriate to include all controls in the accrual model rather than include them as additional independent variables in subsequent empirical analyses. Including variables such as book-to-market and operating performance in the accrual model is beneficial as we are able to identify industry-year specific coefficients. Including these variables as controls in subsequent tests examining the relation between abnormal accruals (without book-to-market and operating performance) and audit fees fails to capture these potentially important industry-year effects.
14
The model in equation (1) was estimated using all firm-years from 1988-2001 on
Compustat that have the required information to calculate an estimate of total accruals
from the statement of cash flows along with estimates of property, plant and equipment
and book-to-market. The model was estimated individually for each two digit SIC code
with at least 8 observations, or a total of 648 industry-year regressions. The mean
coefficient estimates for the parameters for our model and distributional properties of the
resulting accrual measures are presented in Table 3.
Consistent with prior research we find a positive coefficient on (∆Sales-∆REC)
and a negative coefficient on PPE (the traditional parameters in the modified Jones
model). We also find that BM and CFO are both negatively associated with total
accruals. The mean adjusted R2 is about 30 percent in our extended accrual model.
For the pooled sample of 62,766 firm-year observations used to generate our
measures of unexpected accruals, the mean value of unexpected accruals is zero by
construction (unexpected accruals are the residual from a regression model). Hence,
finding zero unexpected accruals suggests we have random sample drawn from the
Compustat population.
In the tests that follow, we examine both the raw values for abnormal accruals
along with their absolute values.3 However, there is little consensus on the precise
accrual measure for earnings quality. If the earnings management is directional, the
appropriate metric is the raw value. For example, if payments to auditors creates an
incentive to engage in income increasing earnings management then the research design
3 In unreported results we also replicated all results using alternative measures of abnormal accruals using a more limited set of independent variables. Our conclusions are unaffected by these alternate measures of abnormal accruals.
15
should focus on signed measures of unexpected accruals. However, if the earnings
management is non-directional, the absolute value of unexpected accruals is appropriate.
For example, if payments to auditors are expected to create incentives to engage in
income increasing and decreasing behavior then the research design should focus on
deviations from a “normal” level of total accruals.
Previous research has found that the provision of non-audit services is associated
with higher absolute values of unexpected accruals (Frankel et al., 2002). Klein (2002)
also finds that the presence of independent outsiders on the board is associated with
smaller absolute values of unexpected accruals. However, it is not clear whether these
results are robust to directional measures of unexpected accruals. Ashbaugh et al. (2003)
find some evidence of a positive relation between the provision of non-audit services and
directional measures of unexpected accruals. However, they find that firms with negative
unexpected accruals drive this relation. Since the measurement of accruals is
controversial, we include both directional and absolute measures of the respective accrual
variables. In addition, we also provide separate analyses for firms with positive and
negative accruals in order to determine whether incentives for earnings management vary
depending on the sign of the accrual.
3.4 Measurement of Corporate Governance
We obtain data on corporate governance for our sample from a variety of sources.
Ownership data by institutions and insiders were obtained from SEC disclosures that
identify shareholdings of corporate insiders and all institutional fund managers who hold
at least $100 million in exchange listed or NASDAQ quoted securities at the end of the
year. These disclosures are made on SEC form 13F, which describes the number of
shares and market value of each security held. These data items were obtained from
16
WorldScope. Data on board characteristics were obtained from Institutional Shareholder
Services (ISS). ISS classifies each director on the board as an (i) insider, (ii) affiliated, or
(iii) independent outsider. We measure Board Composition as the number of
independent outsiders on the board divided by Board Size.
The descriptive statistics for the governance variables are presented in Table 4.
Because our governance data are drawn from several sources we do not require complete
data availability for all measures to maximize the sample size. We have data available on
insider and institutional holdings for 2,742 of our 3,424 firms. Data are available for
board composition for 1,986 firm observations. The mean (median) firm in our sample
has approximately 44 (43) percent of its outstanding shares held by institutions and 16 (8)
percent held by corporate insiders. The mean (median) percentage of outside directors on
the board is 60 (62) percent. Table 4 (Panel B) reports the correlations between our
selected governance measures. Pearson (Spearman) correlations are reported below
(above) the diagonal and correlations. There is a strong negative correlation between
institutional holdings and holdings by corporate insiders. Institutional holdings are
greater for firms with larger and more independent boards. These governance data are
consistent with descriptive statistics reported in previous research (e.g., Klein, 2002).
Prior research suggests that corporate governance is enhanced by higher
institutional holdings (Shleifer and Vishny, 1997 describe the external monitoring
benefits) and a higher proportion of independent directors (internal monitoring). A large
stream of research has documented some benefit to the presence of independent outsiders
on the board. Directors have a fiduciary duty to exercise care in monitoring management
on behalf of shareholders (Fama and Jensen, 1983). A firm with an outside director
majority is more likely to replace a CEO following poor firm performance (Weisbach,
17
1988), make better acquisitions (Byrd and Hickman, 1992), adopt poison pills to improve
shareholder value (Brickley et al., 1994) and have higher quality reported earnings
(Dechow et al., 1996).
A larger proportion of insider holdings can also improve corporate governance if
standard agency problems between stockholders and managers are mitigated. However,
as insider holdings increase, it is possible that insiders de facto assume control of the
firm, and this exacerbates agency problems (Morck et al., 1988, describe how insiders
can exploit the firm for intermediate levels of ownership).
3.5 Econometric Approach
A common assumption that is made in the prior research is that one structural
model is appropriate for the entire sample. However, if different models characterize
subsets of observations, the pooled estimation results can be highly misleading.4 An
alternative to using a pooled estimation approach is to classify the sample into
homogenous clusters of observations with similar regression models. Moreover, once
these clusters are identified, it is possible to determine what distinguishing factors are
associated with the observations in each cluster. For example, payment for non-audit
services can have positive relation with accrual behavior for some firms, negative for
some firms, and zero for still other firms. Given these clusters, standard statistical
methods can be used determine whether cluster membership (which corresponds to
different structural models) is related to differences in corporate governance.
4 Although a somewhat extreme example, it is possible for the regression model for 50% of the sample to have a positive regression coefficient and the regression model for the remaining 50% of the sample to have a negative coefficient with the same absolute value. Under this scenario, the estimated regression coefficients for the pooled sample (assuming equal error variances) will be zero. Clearly, this erroneous interpretation occurs because of an inappropriate pooling of the data.
18
There is an extensive literature in statistics, econometrics, and marketing that
provides the methodological basis for univariate mixtures of distributions (e.g.,
Newcomb, 1886 and Pearson, 1894) and switching regressions (e.g., Quandt and
Ramsey, 1978). In addition, DeSarbo and Cron (1988), Wedel and DeSarbo (1995), and
others have developed maximum likelihood methods for clusterwise regressions or latent
class mixture models. We employ latent class models in our empirical analysis.
In formal terms, the latent class model (or finite mixture model) assumes that the
sample of N observations can be characterized with K clusters. Each observation is yi =
XiΒ + ει, where yi is the dependent variable for the ith observation, Xi is the set of
independent variables for the ith observation, Β is the corresponding set of regression
coefficients for the K clusters, and ει is the normally distributed error term for the
regression model.5 If yi is distributed as a finite mixture of normal distributions, a sample
of N independent observations produces the following likelihood expression:
∏
σ
Β−−πσ∑ λ=
=
−
=
N
1i2k
2kii2/1
k
K
1kk ]
2)Xy(
exp[)2(L , (2)
where λk is the unknown proportion of the sample that is contained in cluster k and σk is
the standard deviation of the error term in cluster k. The unknown parameters are λk, σk,
and Βk. Estimates of these unknown parameters are obtained by maximizing the above
likelihood function subject to 10 k ≤λ≤ , 1K
1kk =∑ λ
=, and 2
kσ > 0.
One difficulty in applying latent class mixture models is the determination of the
number of clusters (i.e., K). As with many multivariate procedures, formal statistics for
5 See DeSarbo and Cron (1988), Wedel and DeSarbo (1995), and references contained in these papers for the precise technical optimization details of clusterwise regressions and latent class modeling. The statistical discussion in the text is adapted from these papers.
19
this choice are not available and heuristic fit statistics based on the logarithm of the
likelihood function (adjusted for the number of parameters used in the estimation) are
used. As suggested by Akaike (1974) and Bozdogan (1987), a reasonable fit statistic is
the Consistent Akaike Information Criterion (CAIC). The CAIC is equal to
-2 ln L + [P·K + 2K -1]·(ln(N + 1), L is the maximum likelihood for the K cluster
solution, P is the number of independent variables in the regression model, K is the
number of clusters, and N is the sample size. The number of clusters is determined for
that value of K that minimizes the CAIC.
Given the number of clusters and the parameter estimates, the posterior
probability that each observation belongs to a specific cluster can be computed using
Bayes theorem (which is referred to as a “fuzzy” clustering). A natural classification rule
is to assign each observation to the cluster where it has the largest relative posterior
probability. The estimated λk (and the proportion of firms assigned to each cluster) is an
important parameter because it will identify whether a relation between audit fees and
accrual behavior characterizes the entire sample (i.e., λ = 1) or a smaller subset of the
observations (i.e., λ < 1). After firms are assigned to one specific cluster, the differences
in corporate governance across clusters (where a different relation between accruals and
payments for non-audit services characterizes each cluster) can be completed using
standard generalized linear model methods.
Rather than using latent class mixture modeling, prior research estimates a single
(pooled) regression model and then the model is extended to incorporate (usually
exploratory) interactions for subsets of interest (e.g., ownership variables). Although the
use of interactions (e.g., fees • institutional ownership) can provide useful insights into
the conditional relation between accruals and audit fees, this approach has several
20
inherent limitations. First, there are numerous governance variables of interest and the
analysis will require many interactions in the regression model. These numerous
interactions will almost certainly result in high levels of multicollinearity, and this will
make the interpretation of the statistical significance for the coefficients problematic
(e.g., Yi, 1989). Second, typical interaction models allow the slope terms to vary, but
constrain the intercept (the conditional mean level for accruals) to be identical for all
firms in the pooled sample. However, Klein (2002), Xie et al. (2002) and Jenkins (2002)
find that the level of accruals varies with respect to selected governance variables. Thus,
it is questionable whether the implicit constraint on the intercept is appropriate.6 Finally,
interactions between continuous variables assume that the incremental impact on the
slope is either always increasing or decreasing. Unfortunately, if the impact of a
governance variable on the slope is nonlinear, the structure imposed by the interaction
will be inappropriate.7
Latent class mixture models do not impose this type of interaction structure on the
relation between accruals and audit fees. In contrast, this approach explicitly allows for
the possibility that there are different models linking accruals to audit fees (involving a
different intercept and/or slope). Once clusters of firms with a homogenous relation
between accruals and audit fees are identified, it is then possible to uncover whether
governance factors differ among the clusters. Although our approach has a very different
theoretical orientation than pooled regression models with interactions, interaction
6 It is, of course, possible to have a separate intercept for each group of firms using interactions. However, the groups (or clusters) of firms must be known a priori in order to implement this approach. Unfortunately, researchers do not typically have such strong a priori knowledge. 7 One solution to this problem is to shift from interactions between continuous variables to an interaction between audit fees and a categorical governance variable (e.g., high, medium and low institutional ownership). This type of interaction can uncover nonlinear features in the data. However, the conversion of a continuous variable into a categorical variable is almost completely arbitrary and it is not clear how to assess model misspecification.
21
structures are simply special cases of latent class mixture models. Given the controversy
surrounding the actual relation (if any) between audit fees and earnings quality, a more
general analysis approach is an appropriate choice.
4. Results
4.1 Pooled Sample
In order to link our results to prior research, we first report the estimation results
for each of our accrual and payment measures with the basic regression model:
Accrual_Measurei = α+ β Audit Fee Measurei + εi (3)
Consistent with Frankel et al. (2002), RATIO has a marginally statistically positive
association with the non-directional accrual (Table 5, Panel B), but no statistical
association with any of the other accrual measures. TOTFEE and NONAUDFEE have a
statistically positive relation with directional accruals, which suggests that accruals are
higher in firms that pay larger fees to their auditors. However, TOTFEE and
NONAUDFEE have a statistically negative relation with non-directional accruals,
negative relation when accruals are restricted to values greater than zero, and positive
relation when accruals are restricted to values less than or equal to zero. These results
indicate that firms making larger payments to their auditors exhibit smaller accruals.
These contradictory results are likely to be caused by the extremely low level of
explanatory power of the regressions in Table 5. The weak statistical and substantive
relation for the pooled sample results is consistent with the conflicting evidence reported
in papers such as Frankel et al. (2002), Antle et al. (2002) and Ashbaugh et al. (2003).
One important assumption underlying the results in Table 5 is that a single
regression model is appropriate for the entire sample of firms. That is, similar to prior
22
research, the results assume that all of the sample observations are homogenous with
respect to the relation between the fees paid to auditors and accrual activity. In the next
section, we relax this assumption and analyze the same set of data using latent class
mixture models.
4.2 Latent Class Mixture Analysis
The regression model in equation (3) was estimated using a latent class mixture
approach for each combination of accrual and audit fee measures. The minimum CAIC
statistic (unreported) was achieved with three clusters for each combination. The
parameter estimates, t-statistics, and proportion of the sample that is contained in each of
the three clusters are presented in Tables 6, 7 and 8 for RATIO, TOTFEE and
NONAUDFEE, respectively.
There is no statistical association with between RATIO and directional accruals or
accruals that are constrained to be positive (Table 6, Panels A and C).8 However, Cluster
I exhibits a positive relation with non-directional accruals (Panel B) and a negative
relation with accruals that are constrained to be less than or equal to zero (Panel D).
Although these results are consistent with Frankel et al. (2002), it is important to note that
this association only applies to 8.5 percent of the sample in Panel B and 14.8 percent of
the sample in Panel D. Thus, if RATIO is a valid measure for describing the relation
between the auditor and a client, accounting quality is only sacrificed for a small subset
of firms.
As discussed in Section 3.4, the estimation results can be used to compute the
posterior probability that each observation belongs to a specific cluster. Based on these
23
posterior probabilities, we assign each observation to the cluster where it has the largest
estimated posterior probability. Given these classifications, it is possible to use standard
analysis of variance (ANOVA) methods to determine whether there are differences in
corporate governance and other variables across the three clusters.9
In Table 6 (Panels B and D), we find that firms in the cluster exhibiting a positive
relation between abnormal accruals and RATIO have smaller book-to-market ratios (i.e.,
high growth prospects) and market capitalizations than the remaining sample firms. In
addition, these firms have lower institutional and higher insider holdings. These results
are consistent with a firm that is difficult to monitor and where insiders have effective
control of the organization. Thus, concurrent weakness in corporate governance appears
to be an important determinant of the relation between auditor independence and earnings
quality.10
The results for TOTFEE and NONAUDFEE are presented in Tables 7 and 8,
respectively. Given the substantial correlation between these two measures (Table 2,
Panel B), it is not surprising that the results for TOTFEE and NONAUDFEE are virtually
identical. There are no statistically significant results for the slope coefficients when
directional accruals are used as the measure for earnings quality. However, large and
statistically significant results are observed with the remaining accrual measures. The
slope coefficient (β) is consistently negative for the non-directional accruals, negative for
8 In the analyses where the slope term is not statistically significant, the clusters differ only in terms of the intercept (which is equivalent to examining differences in the mean level of accrual across clusters). 9 In each table we report an F-test that tests the null hypothesis of no difference across all three clusters. In unreported tests we have also performed pair-wise tests using a Tukey HSD test. Our discussion of significant differences is based on these pair-wise differences. 10 Prior research by Frankel et al (2002) and Ashbaugh et al. (2003) documents that firms with greater institutional holdings report smaller accruals. These papers, however, examine institutional holdings as a control variable and not as part of the overall governance structure that impacts the relation between
24
the accruals that are constrained to be greater than zero, and positive for accruals that are
constrained to be less than or equal to zero. These results indicate that accruals are
smallest for firms where the auditor has the greatest financial interest. This result is
completely inconsistent with the very modest results obtained using RATIO. The
economic significance of this relation is not trivial. For example, using the estimates for
cluster I from panel B of table 7 a change in TOTFEE equal to its inter-quartile range
(0.021) corresponds to a 0.0034 reduction in the value of |Accruals|. The median firm in
our sample reports operating earnings equal to six percent of its asset base. A 0.0034
reduction in accruals translates to a 6 percent (0.0034/0.06) reduction in ROA.
The largest statistically significant slope coefficient is observed for cluster of
firms that has a relatively low book-to-market ratio (i.e., high growth prospects), low
market capitalization, fewer external board members, low institutional holdings, and high
insider holdings. Thus, the relation between payments to auditors and earnings quality is
the most stringent when the client has weak governance. Overall, our results are most
consistent with reputation concerns being the primary determinant of auditor behavior
with respect to limiting unusual accounting choices of client firms and inconsistent with
payments to auditors causing a decrease in accounting quality.
4.3 Analysis of “Abnormal” Fees
One concern with the use of total and non-audit fees is that we use the overall
level of fees as the measure for the importance of the client to the auditor without any
controls for the source of these fees. In an attempt to address this limitation, we develop
two additional measures of client importance that focus on “abnormal” non-audit fees and
accrual measures and the provision of non-audit services. We are not aware of prior research that has examined the provision of non-audit services as part of a larger governance issue.
25
“abnormal” total fees. Our estimates of the abnormal fees are generated from the
following regression specification (see Simunic 1984 and Craswell et al., 1995 for model
details):11
Log(Fee) = φ0 + φ1Log(Assets) + φ2Log(Segments) + φ3Inventory + φ4Receivables + φ5Debt + φ6Income + φ7LOSS + φ8Opinion + ε (4)
The variables are defined as follows. Log(Fee) is the natural logarithm of either
total fees or non-audit fees paid to the auditor. Log(Assets) is the natural logarithm of
total assets (Compustat data item 6). Log(Segments) is the natural logarithm of the
number of business segments reported on the Compustat Segment Data File. Inventory is
the ratio of the dollar value of inventory (item 3) to total assets (item 6). Receivables are
the ratio of the dollar value of accounts receivable (item 2) to total assets (item 6). Debt
is the sum of short term debt (item 34) and long term debt (item 9) to total assets (item 6).
Income is the ratio of operating income after depreciation (item 178) to average total
assets (item 6). LOSS is coded as an indicator variable that is equal to one if the firm
reports negative Income in any of the previous three years and zero otherwise. Opinion is
an indicator variable equal to one if the firm receives a qualified audit opinion and zero
otherwise. A qualified audit opinion is defined as anything other than the standard
unqualified audit opinion coded as one by Compustat.
Equation (4) is estimated for each of the 54 industry groups (two-digit standard
industrial classification) in our sample. The estimated residual ( ε̂ ) from equation (4) is
our proxy measure for “abnormal” fees. To transform this to a dollar amount we raise
exp to the power of the predicted value of Log(Fee) and then subtract this value from the
11 The model in equation (4) has been developed for audit fees, and not non-audit fees. To our knowledge, prior research has not developed a model for non-audit fees. In the absence of such a model, we simply apply the same set of independent variables to total and non-audit fees. We acknowledge that our analysis
26
dollar fee. The result (after deflating by auditor firm revenue) is denoted as ABTOTFEE
(ABNONAUDFEE) for total fees and non-audit fees.
The explanatory power from this specification for fees is very high with the mean
adjusted R2 of about 75 percent for the total fee model and 60 percent for the non-audit
fee model (Table 9). The mean coefficient estimate and mean t-statistics across the 54
industry groups is also presented in Table 9.12 As expected, fees are positively associated
with firm size, number of segments (a measure of audit complexity), extent of inventory
(a measure of audit complexity), existence of a loss and a qualified opinion (which
requires more audit effort). In general, the estimated coefficients and explanatory power
of our model are very similar to prior research (e.g., Simunic 1984), with firm size being
the key independent variable.
We expect auditor behavior to vary depending on whether the auditor is being
paid more or less than the economic benchmark for a specific client.13 When the
“abnormal” fee is less than or equal to zero (denoted as either LowABTOTFEE or
LowABNONAUDFEE), the auditor has little to lose if they impose stringent accounting
requirements on the client that result in lower levels of accruals. If this action causes the
auditor to lose this client, we assume that there are other more profitable uses for the staff
previously assigned to this client. However, if the client remains with the auditor, there is
considerable incentive for the auditor to be aggressive with this client in order to
minimize any reputation loss due to an audit failure. Obviously, the case where the
of non-audit fees may be confounded by the use of an inappropriate model for computing “abnormal” non-audit fees. 12 We acknowledge that the 54 regressions are likely to exhibit positive cross-sectional correlation. Although we do not know the extent of cross-sectional correlation, the z-statistic that results from the aggregation of individual t-statistics will range from the mean t-statistic (if the correlation is equal to one) to approximately 54 times the mean t-statistic (if the correlation is equal to zero). The latter computation is labeled as the maximum t-statistic in Table 9.
27
auditor would be most susceptible to client pressure is when the “abnormal” fee is greater
than zero (denoted as either HighABTOTFEE or HighABNONAUDFEE). If reputation
concerns are not relevant to the auditor, we should observe lower earnings quality for
firms that pay a positive premium to the auditor for audit and non-audit work.
The results of the latent class mixture analysis using the “abnormal” fee measures
are presented in Table 10. For the “abnormal” total fee measures (Panels A to D), there
are no statistically significant slope coefficients for directional accruals. However, we
find consistent results across the three clusters for the other three accrual measures.
Positive “abnormal” fees are associated with lower non-directional accruals and smaller
(closer to zero) negative and positive accruals. The same results are observed with
negative “abnormal” fees (after incorporating the necessary shift in the sign of the
coefficient). In addition, we find virtually the same results with “abnormal” non-audit
fees. The only difference is that there is a statistically positive coefficient on
LowABNONAUDFEE the directional accrual measure (which indicates that as the
“abnormal” fee becomes more negative, the accrual becomes more negative). Overall,
the results in Table 10 are consistent with reputation being an important determinant of
earnings quality. We find no evidence that large “abnormal” payments to auditors cause
a decline in earnings quality. In fact, our results indicate precisely the opposite outcome.
5. Summary and Conclusions
The relation between the fees paid to auditors and earnings quality has been the
focus of considerable scholarly, institutional and regulatory debate. Assumptions and
conjectures about this relationship have been instrumental in shaping the regulatory
13 In a recent paper Raghunandan, Read and Whisenant (2003) find no evidence of abnormally high fee payments to restatement firms compared to a matched sample.
28
debate concerning the forced divestiture of the consulting function in audit firms and the
general structure for the auditing industry. Unfortunately, prior research by Frankel et al.
(2002), Antle et al. (2002) and Ashbaugh et al. (2003) have found mixed results on this
important empirical association. Therefore, the role of fees for audit and non-audit
services on accounting choices is an unresolved issue.
Similar to some prior research, we find very little evidence of a positive relation
between the fees paid to auditors and measures of accruals for a large pooled sample of
firms. We find a positive association between audit fees and unexpected accruals only
when audit fees are measured using the ratio of audit fees to total fees paid to the auditor
and unexpected accruals are transformed using the absolute value (i.e., a non-directional
measure). Moreover, the statistically positive association between non-audit fees and
accrual behavior only occurs for approximately 8.5 percent of the total sample. This
small cluster of firms, relative to the remaining clusters, has a smaller market
capitalization, lower book-to-market ratio, lower institutional holdings, and higher insider
holdings. Thus, corporate governance is a key factor for understanding accrual choices,
as opposed to these choices simply being a function of fees paid to auditors.
Using alternate measures of audit fees which capture the extent of financial
dependence of a given client to an auditor, we find a statistically negative relation
between auditor independence and earnings quality. The cluster of firms with the most
pronounced negative association is characterized by low market capitalization, high
growth prospects, less independent boards, low institutional holdings, and high insider
holdings. For these firms the auditor appears to be playing a key role in the governance
process to limit abnormal accrual choices. Similar to Francis and Reynolds (2001), our
29
results are consistent with reputation concerns being the primary determinant of auditor
behavior with respect to limiting unusual accounting choices of client firms.
Our study has several limitations, and it is important to make these problems
explicit. First, auditor choice, purchase of non-audit services, and the governance
structure are endogenous variables. This endogeneity is ignored in our analysis and our
results are subject to the traditional econometric problems caused by endogeneity. With
the exception of the structural modeling approach in Antle et al. (2002), this limitation is
inherent in all prior research examining the relation between non-audit services and
accrual behavior. It is important for future research to develop a more complete set of
structural models with a sophisticated selection of exogenous (or instrumental) variables.
Second, our results are based only on two years of data, and this limits the ability
to generalize the results to other time periods. With the exception of the research by
Kinney et al. (2003) that uses proprietary data, and data from other countries (e.g., U.K.
data used in Antle et al., 2002 and Australian data used in Ruddock et al., 2003), this is
also a limitation to prior research. Ignoring issues of self-selection, the data used by
Kinney et al. are especially intriguing because it is collected during the time period prior
to the auditing controversies. In contrast, the self-reported fee data for the time periods
used in this study may be “manipulated” by the firms in response to shareholder and
legislative inquiries.
Finally, similar to prior research, we use accrual measures as indicators for
earnings quality or “bad behavior” by managers. Measures of unexpected accruals are
criticized because they incorrectly classify “expected” accruals as unexpected. Hence,
any association between payments to auditors and measures of unexpected accruals could
be due to measurement error in unexpected accruals and not “bad managerial behavior.”
30
To help mitigate this problem we employ a model of unexpected accruals that has greater
explanatory power and is less likely to misclassify expected accruals as unexpected
(Dechow et al., 2003). Nevertheless, there is in an unknown degree of measurement error
inherent in our accrual metrics.
31
References Akaike, H. 1974. A New Look at Statistical Model Identification. IEEE Transactions on
Automatic Control 6: 716-723. Antle, Rick, Elizabeth, A. Gordon, Ganapathi Narayanamoorthy, and Ling Zhui. 2002.
The joint determination of audit fees, non-audit fees and abnormal accruals. Working paper, Yale University.
Ashbaugh, Hollis, Ryan LaFond, and Brian W. Mayhew. 2003. Do non-audit services
compromise auditor independence? The Accounting Review, 78, 611-639. Bartov, E., F. A. Gul, and J. S. L. Tsui. 2001. Discretionary accruals models and audit
qualifications. Journal of Accounting and Economics, 30, 421-452. Becker, C., M. DeFond, J. Jiambalvo and K. Subramanyam. 1998. The effect of audit
quality on earnings management. Contemporary Accounting Research, 15, 1-24. Bernard, V. L., and D. J. Skinner. 1996. What motivates managers’ choice of
discretionary accruals? Journal of Accounting and Economics, 22, 313-325. Bozdogan, H. 1987. Model Selection and Akaike’s Information Criterion (AIC): The
General Theory and Its Analytical Extensions. Psychometrika 52: 163-180. Brickley, J., J. Coles and R. Terry, 1994, Outside Directors and the Adoption of Poison
Pills. Journal of Financial Economics, 35, 371-390. Byrd, J. and K. Hickman, 1992, Do Outside Directors Monitor Managers? Evidence
from Tender Offer Bids. Journal of Financial Economics, 32, 195-207. Chung, Hyeesoo and Sanjay Kallapur. 2003. Client importance, non-audit services and
abnormal accruals. Working paper, Purdue University. Craswell, A. T., J. R. Francis and S. L. Taylor. 1995. Auditor brand name reputations
and industry specialization. Journal of Accounting and Economics, 29, 297-322. DeAngelo, L. E. 1981. Auditor size and audit quality. Journal of Accounting and
Economics, 3, 113-127. Dechow, Patricia M., Scott A Richardson, and A. Irem Tuna. 2003. Why are earnings
kinky?: An examination of the earnings management explanation. Review of Accounting Studies, 8, 355-384.
Dechow, Patricia M., Richard G. Sloan, and Amy P. Sweeney. 1995. Detecting Earnings
Management. The Accounting Review, 70, 193-225.
32
Dechow, P. M., R. G. Sloan and A. P. Sweeney. 1996. Causes and Consequences of Earnings Manipulation: An Analysis of Firms Subject to Enforcement Actions by the SEC. Contemporary Accounting Research, 13 (1), 1-36.
Dechow, Patricia M., and Douglas J. Skinner. 2000. Earnings management: Reconciling
the views of accounting academics, practitioners and regulators. Accounting Horizons, 14, 235-250.
DeFond, Mark. L., K. Raghunandan, and K. R. Subramanyam. 2002. Do non-audit
service fees impair auditor independence? Evidence from going concern audit opnions. Journal of Accounting Research, 40, 1247-1274.
DeFond, Mark. L., and K. R. Subramanyam. 1998. Auditor Changes and Discretionary
Accruals. Journal of Accounting and Economics, 25, 35-67. DeSarbo, W. and W. Cron. 1988. A Maximum Likelihood Methodology for Clusterwise
Linear Regression. Journal of Classification 5: 249-282. Fama, E. F. and M. Jensen, 1983, Separation of Ownership and Control. Journal of Law
and Economics,26, 301-325. Francis, Jere, and Bin Ke. 2003. Do fees paid to auditors increase a company’s
likelihood of meeting analysts’ forecasts? Working paper, University of Missouri.
Frankel, Richard M., Marilyn F. Johnson, and Karen K. Nelson. 2002. The relation
between auditor’s fees for non-audit services and earnings management. The Accounting Review, 77, Supplement, 71-105.
Jenkins, Nicole T. 2002. Auditor independence, audit committee effectiveness and
earnings management. Working paper, Washington University. Jones, Jennifer. 1991. Earnings Management During Import Relief Investigations.
Journal of Accounting Research, 29, 193-228. Kinney, William R., Palmrose, Zoe-Vanna, and Susan W. Scholz. 2003. Auditor
Independence and Non-Audit Services: What do Restatements Suggest? Working paper, University of Texas at Austin.
Klein, April. 2002. Audit committee, board of director characteristics, and earnings
management. Journal of Accounting and Economics, 33, 375-400. Magee, R. P., and M-C Tseng. 1990. Audit pricing and independence. The Accounting
Review, 65, 315-336. McNichols, Maureen F. 2000. “Research Design Issues in Earnings Management
Studies. Journal of Accounting and Public Policy, 19, 313-345.
33
McNichols, Maureen F. 2002. Discussion of the quality of accruals and earnings: The role of accrual estimation errors. The Accounting Review, 77, Supplement, 61-69.
Morck, R., A. Shleifer, and R. W. Vishny, 1988. Management ownership and market
valuation: An empirical analysis, Journal of Financial Economics, 20, 293-315. Myers, James N., Linda A. Myers, and Thomas C. Omer. 2003. Exploring the term of
the auditor-client relationship and the quality of earnings: A case for mandatory auditor rotation? The Accounting Review. 78, 779-799.
Newcomb, S. 1886. A Generalized Theory of the Combination of Observations So As to
Obtain the Best Result. American Journal of Mathematics 8: 343-366. Palmrose, Zoe-Vonna. 1986. The effect of non-audit services on the pricing of audit
services: Further evidence. Journal of Accounting Research, 24, 405-411. Pearson, K. 1894. Contributions to the Mathematical Theory of Evolution. Philosophical
Transactions A. 185: 71-110. Quandt, R. and J. Ramsey. 1978. Estimating Mixtures of Normal Distributions and
Switching Regressions. Journal of the American Statistical Association. 73: 730-738.
Raghunandan, K., W. J. Read, and J. S. Whisenant. 2003. Initial evidence on the
association between nonaudit fees and restated financial statements. Accounting Horizons, 17, 223-234.
Reynolds, J. and J. Francis. 2001. Does size matter? The influence of large clients on
office level auditor reporting decisions. Journal of Accounting and Economics, 30, 375-400.
Ruddock, Caitlin, Sarah Taylor, and Stephen Taylor. 2003. Non-audit services and
earnings conservatism: Is auditor independence impaired? Working paper, University of Technology, Sydney.
Shleifer, A. and R. W. Vishny, 1997, A Survey of Corporate Governance. Journal of
Finance, 52 (2), 737-783. Simunic, Dan A. 1984. Auditing, consulting, and auditor independence. Journal of
Accounting Research, 22, 679-702. Wedel, M. and W. DeSarbo. 1995. A Mixture Likelihood Approach for Generalized
Linear Models. Journal of Classification 12: 21-55. Weisbach, M., 1988, Outside Directors and CEO Turnover. Journal of Financial
Economics, 20, 431-460.
34
Xie, Biao, Wallace N. Davidson III, and Peter J Dadalt. 2002. Earnings management and corporate governance: The roles of the board and the audit committee. Working paper, Southern Illinois University.
Yi, Y., 1989. On the evaluation of main effects in multiplicative regression models.
Journal of the Market Research Society, 31, 133-138.
Table 1
Descriptive Statistics for the Sample
The sample includes 5,103 firm-year observations for the fiscal years 2000 and 2001 for which we are able to calculate measures of abnormal accruals and non-audit services.
Panel A: Industry Composition
2 digit SIC Industry Number Percent of Sample
Compustat Composition
1 Crops 13 0.25 0.29 10 Ores 20 0.39 1.12 13 Oil & Gas 160 3.14 3.61 14 Quarry 13 0.25 0.29 15 Building - Light 26 0.51 0.46 16 Building - Heavy 16 0.31 0.34 17 Construction 15 0.29 0.32 20 Food 101 1.98 2.26 22 Textile Mill 22 0.43 0.40 23 Apparel 47 0.92 0.93 24 Lumber 30 0.59 0.51 25 Furniture 37 0.73 0.53 26 Paper 60 1.18 1.04 27 Printing 67 1.31 1.33 28 Chemicals 470 9.21 8.56 29 Petroleum 21 0.41 0.62 30 Rubber 56 1.10 1.17 31 Leather 18 0.35 0.36 32 Stone 35 0.69 0.67 33 Metal Work - Basic 89 1.74 1.52 34 Metal Work – Fabrication 87 1.70 1.44 35 Industrial 357 7.00 6.84 36 Electrical 479 9.39 8.88 37 Transport – Equipment 116 2.27 1.79 38 Instruments 372 7.29 7.03 39 Misc. Manufacturing 47 0.92 1.14 40 Railroad 12 0.24 0.27 42 Motor freight 44 0.86 0.72 44 Water Transport 27 0.53 0.44 45 Air Transport 42 0.82 0.57 47 Transport - Services 19 0.37 0.35 48 Communications 151 2.96 4.22 49 Utilities 187 3.66 3.08 50 Durables – wholesale 137 2.68 2.54 51 NonDurables - wholesale 77 1.51 1.58 52 Garden 15 0.29 0.20 53 General Stores 42 0.82 0.58 54 Food Stores 26 0.51 0.56
Table 1 (continued)
2 digit SIC Industry Number Percent of Sample
Compustat Composition
55 Auto Dealers 30 0.59 0.44 56 Apparel - Retail 72 1.41 0.88 57 Home Equipment 34 0.67 0.45 58 Eating 71 1.39 1.52 59 Misc. Retail 103 2.02 2.08 70 Hotels 24 0.47 0.48 72 Personal Services 17 0.33 0.31 73 Business Services 840 16.46 17.69 75 Auto Repair 7 0.14 0.24 78 Movies 14 0.27 0.65 79 Amusements 56 1.10 1.15 80 Health 88 1.72 1.72 82 Educational 21 0.41 0.42 83 Social 15 0.29 0.26 87 Engineering – Retail 147 2.88 2.46 99 Nonclassifiable 11 0.22 0.70
Panel B: Distributional statistics for firm characteristics
Variable Mean Std. Dev. Q1 Median Q3
Earnings -0.048 0.229 -0.097 0.020 0.072
CFO 0.041 0.196 -0.014 0.070 0.140
Book-to-Market 0.784 0.856 0.283 0.520 0.944
Log(Market Value) 5.641 2.038 4.240 5.602 6.933
CFO is the operating cash flows (item 308). Earnings is income before extraordinary items (item 123) as reported on the statement of cash flows. Book-to-Market (BM) is the book to market ratio calculated as the book value of common equity (item 60) divided by market capitalization at the end of the fiscal year (item 25 * item 199). CFO and Earnings are scaled by average total assets using assets from the start and end of the fiscal year.
Table 2
Descriptive statistics for the audit fee variables.
Analysis of various audit fee measures for our sample of 5,103 firm year observations
with available Compustat and Standard and Poors audit fee data in 2000 and 2001.
Panel A: Distributional statistics Variable Mean Std. Dev. Q1 Median Q3
RATIO 0.483 0.222 0.317 0.496 0.660
TOTFEE 0.032 0.067 0.005 0.011 0.026
NONAUDFEE 0.020 0.047 0.002 0.005 0.015
Panel B: Correlations Audit Fee Variable RATIO TOTFEE NONAUDFEE
RATIO - 0.550** 0.764**
TOTFEE 0.321** - 0.940**
NONAUDFEE 0.391** 0.978** -
** (*) Significant at the 1% (5%) level. Pearson correlation coefficients are reported below the diagonal and Spearman above. RATIO is the ratio of fees paid to auditors for non-audit services divided by the total fees (the sum of audit and non-audit fees) paid to auditors. TOTFEE is the ratio of total fees (the sum of audit and non-audit fees) paid to the auditor, to the total revenue received that year by the auditor. NONAUDFEE is the ratio of non-audit fees paid to the auditor, to the total revenue received that year by the auditor.
Table 3
Analysis of model of abnormal accruals using data from 1988-2001.
Panel A: Mean coefficient estimates for accrual models based on 648 two-digit SIC-year regressions
Independent Variables ∆Sales-∆REC PPE BM CFO Adj. R2
0.060 (11.62)
-0.014 (-3.40)
-0.010 (-5.63)
-0.376 (-31.71)
0.301
Panel B: Adjusted R2 Across 648 industry-year regressions
Distribution Statistics Mean Std. Dev. Q1 Median Q3
0.301 0.249 0.109 0.256 0.477
Panel C: Distributional statistics for 5,103 firm year observations
Variable Mean Std. Dev. Q1 Median Q3
Accruals 0.00 0.13 -0.03 0.01 0.06
|Accruals| 0.08 0.10 0.02 0.05 0.10
Parameter estimates are averages from the respective 648 two-digit SIC-year regressions. T-statistics are reported in parentheses below parameter estimates. Standard errors are based on the distribution of two-digit SIC-year parameter estimates. The accrual model is estimated using the Jones (1991) technique of decomposing total accruals into a normal (expected) and abnormal (unexpected) component. The method of decomposition is as follows:
TA = α + β1(∆Sales-∆REC) + β2PPE + β3BM + β4CFO + ε (1) TA is the difference between operating cash flows (item 308) and income before extraordinary items (item 123) as reported on the statement of cash flows. ∆Sales is the change in sales (item 12) for the year. ∆REC is the change in receivables reported on the statement of cash flows (item 302) for the year. PPE is the gross amount of property, plant and equipment (item 7). CFO is the operating cash flows (item 308). Book-to-Market (BM) is the book to market ratio calculated as the book value of common equity (item 60) divided by market capitalization at the end of the fiscal year (item 25 * item 199). Accruals is the residual from equation (1) above, |Accruals| is the absolute value of the residual from equation (1) above. This is our estimate of abnormal accruals. All variables used in the abnormal accrual model (except BM) are scaled by average total assets using assets from the start and end of the fiscal year (item 6).
Table 4
Descriptive statistics for the governance variables.
Analysis of various corporate governance measures for our sample of 5,103 firm year
observations with available Compustat and Audit Fee data in 2000 and 2001. Panel A: Distributional statistics
Variable Mean Std. Dev. Q1 Median Q3
Institutional Holdings 0.436 0.262 0.218 0.427 0.655
Insider Holdings 0.164 0.213 0.012 0.076 0.233
Board Composition 0.601 0.186 0.333 0.615 0.750
Panel B: Correlations
Governance Variable
Institutional Holdings Insider Holdings Board Composition
Institutional Holdings - -0.307** 0.248**
Insider Holdings -0.313** - -0.300**
Board Composition 0.239** -0.272** -
** (*) Significant at the 1% (5%) level. Pearson correlation coefficients are reported below the diagonal and Spearman above. Institutional holdings are the fraction of outstanding shares that are held by institutions (as reported by WorldScope). Insider holdings are the fraction of outstanding shares that are held by insiders (as reported by WorldScope). Board Composition is the fraction of directors serving on the board who are independent from management.
Table 5
Pooled regression analysis.
Analysis of the relation between various audit fee and accrual measures for our sample of 5,103 firm year observations with available Compustat and Audit Fee data in 2000 and
2001.
Accrual Measurei = α + β Audit Fee Measurei + εi (3) Panel A: Directional Accrual Measure (Accruals)
Audit Fee Measure RATIO TOTFEE NONAUDFEE
α 0.002 (0.45)
-0.001 (-0.58)
-0.001 (-0.36)
β -0.002 (-0.29)
0.063 (2.31)
0.077 (1.99)
R2 0.000 0.001 0.001
Panel B: Non-Directional Accrual Measure (|Accruals|)
Audit Fee Measure RATIO TOTFEE NONAUDFEE
α 0.077 (23.20)
0.087 (56.17)
0.086 (56.84)
β 0.011 (1.69)
-0.130 (-6.21)
-0.172 (-5.81)
R2 0.000 0.007 0.006
Panel C: Positive Abnormal Accruals Only (Accruals+)
Audit Fee Measure RATIO TOTFEE NONAUDFEE
α 0.068 (19.26)
0.075 (45.99)
0.075 (46.72)
β 0.010 (1.43)
-0.075 (-3.52)
-0.104 (-3.41)
R2 0.000 0.004 0.004
Panel D: Negative Abnormal Accruals Only (Accruals-)
Audit Fee Measure RATIO TOTFEE NONAUDFEE
α -0.091 (-14.74)
-0.102 (-35.92)
-0.101 (-36.23)
β -0.011 (-0.97)
0.205 (5.09)
0.265 (4.66)
R2 0.000 0.011 0.009
Parameter estimates are for the pooled sample of 5,103 observations form the fiscal year ended 2000 and 2001. T-statistics are reported in parentheses below parameter estimates. RATIO is the ratio of fees paid to auditors for non-audit services divided by the total fees (the sum of audit and non-audit fees) paid to auditors. TOTFEE is the ratio of total fees (the sum of audit and non-audit fees) paid to the auditor, to the total revenue received that year by the auditor. NONAUDFEE is the ratio of non-audit fees paid to the auditor, to the total revenue received that year by the auditor.
TA = α + β1(∆Sales-∆REC) + β2PPE + β3BM + β4CFO + ε (1) TA is the difference between operating cash flows (item 308) and income before extraordinary items (item 123) as reported on the statement of cash flows. ∆Sales is the change in sales (item 12) for the year. ∆REC is the change in receivables reported on the statement of cash flows (item 302) for the year. PPE is the gross amount of property, plant and equipment (item 7). CFO is the operating cash flows (item 308). Book-to-Market (BM) is the book to market ratio calculated as the book value of common equity (item 60) divided by market capitalization at the end of the fiscal year (item 25 * item 199). Accruals is the residual from equation (1) above, |Accruals| is the absolute value of the residual from equation (1) above. Accruals+ is equal to Accruals when Accruals>0 and zero otherwise. Accruals- is equal to Accruals when Accruals<0 and zero otherwise. All variables used in the abnormal accrual model (except BM) are scaled by average total assets using assets from the start and end of the fiscal year.
Table 6
Latent Class Mixture Analysis – Using RATIO.
Panel A: Accruals = α + βRATIO
Cluster α β % Sample Book-to-Market
Log (Market Value)
Board Composition
Institutional Holdings
Insider Holdings
I 0.019 (4.58)
-0.008 (-0.96) 37.1 0.810 5.922 0.610 0.462 0.152
II 0.019 (2.80)
0.006 (0.46) 44.3 0.771 5.462 0.586 0.427 0.175
III -0.079 (-3.63)
0.001 (0.02) 18.6 0.708 4.952 0.607 0.322 0.187
Model R2 = 0.086883 F-test 3.54* 65.52** 6.09** 47.11** 7.85**
Panel B: |Accruals| = α + βRATIO
Cluster α β % Sample Book-to-Market
Log (Market Value)
Board Composition
Institutional Holdings
Insider Holdings
I 0.282 (13.25)
0.081 (2.21) 8.5 0.697 5.025 0.609 0.303 0.186
II 0.126 (17.60)
0.007 (0.61) 21.9 0.736 5.177 0.582 0.394 0.184
III 0.039 (22.08)
0.001 (0.02) 69.6 0.806 5.829 0.605 0.456 0.156
Model R2 = 0.675570 F-test 4.58* 60.15** 3.74* 47.99** 8.56**
Panel C: Accruals+ = α + βRATIO
Cluster α β % Sample Book-to-Market
Log (Market Value)
Board Composition
Institutional Holdings
Insider Holdings
I 0.310 (7.14)
0.061 (0.80) 3.5 0.597 5.310 0.578 0.311 0.179
II 0.122 (14.30)
0.010 (0.71) 24.3 0.708 5.370 0.594 0.412 0.181
III 0.042 (18.47)
-0.001 (-0.01) 72.2 0.797 5.897 0.599 0.470 0.156
Model R2 = 0.617301 F-test 4.16* 18.45** 0.35 16.78** 2.73
Panel D: Accruals- = α + βRATIO
Cluster α β % Sample Book-to-Market
Log (Market Value)
Board Composition
Institutional Holdings
Insider Holdings
I -0.281 (-10.72)
-0.081 (-1.82) 14.8 0.724 4.944 0.613 0.304 0.198
II -0.137 (-10.42)
-0.002 (-0.07) 18.6 0.765 4.839 0.574 0.349 0.186
III -0.034 (-12.25)
-0.001 (-0.17) 66.6 0.812 5.703 0.612 0.434 0.159
Model R2 = 0.709783 F-test 1.63 37.19** 3.73* 26.70** 3.56*
** (*) Significant at the 1% (5%) level. Model R2 is computed as 1 – L/L0 , where L is the maximum for the log-likelihood for the threecluster solution and L0 is log-likelihood for the null model (intercept only).
RATIO is the ratio of fees paid to auditors for non-audit services divided by the total fees (the sum of audit and non-audit fees) paid to auditors. TOTFEE is the ratio of total fees (the sum of audit and non-audit fees) paid to the auditor, to the total revenue received that year by the auditor. NONAUDFEE is the ratio of non-audit fees paid to the auditor, to the total revenue received that year by the auditor. Book-to-Market (BM) is the book to market ratio calculated as the book value of common equity (item 60) divided by market capitalization at the end of the fiscal year (item 25 * item 199). Board Composition is the fraction of directors serving on the board who are independent from management. Institutional Holdings is the fraction of outstanding shares that are held by institutions (as reported by WorldScope). Insider Holdings is the fraction of outstanding shares that are held by insiders (as reported by WorldScope). The accrual model is estimated using the Jones (1991) technique of decomposing total accruals into a normal (expected) and abnormal (unexpected) component. The method of decomposition is as follows: TA = α + β1(∆Sales-∆REC) + β2PPE + β3BM + β4CFO + ε (1) TA is the difference between operating cash flows (item 308) and income before extraordinary items (item 123) as reported on the statement of cash flows. ∆Sales is the change in sales (item 12) for the year. ∆REC is the change in receivables reported on the statement of cash flows (item 302) for the year. PPE is the gross amount of property, plant and equipment (item 7). CFO is the operating cash flows (item 308). All variables used in the abnormal accrual model (except BM) are scaled by average total assets using assets from the start and end of the fiscal year. Accruals is the residual from equation (1) above, |Accruals| is the absolute value of the residual from equation (1) above. Accruals+ is equal to Accruals when Accruals>0 and zero otherwise. Accruals- is equal to Accruals when Accruals<0 and zero otherwise.
Table 7
Latent Class Mixture Analysis – Using TOTFEE Panel A: Accruals = α + βTOTFEE
Cluster α β % Sample Book-to-Market
Log (Market Value)
Board Composition
Institutional Holdings
Insider Holdings
I 0.022 (6.98)
-0.002 (-0.03) 44.7 0.770 5.475 0.588 0.428 0.174
II 0.016 (8.37)
-0.015 (-0.58) 36.6 0.811 5.923 0.609 0.462 0.153
III -0.086 (-8.00)
0.326 (1.35) 18.7 0.709 4.944 0.606 0.322 0.187
Model R2 = 0.090106 F-test 3.71* 65.31** 5.09** 47.17** 7.14**
Panel B: |Accruals| = α + βTOTFEE
Cluster α β % Sample Book-to-Market
Log (Market Value)
Board Composition
Institutional Holdings
Insider Holdings
I 0.135 (33.29)
-0.163 (-4.02) 22.2 0.727 5.218 0.583 0.397 0.186
II 0.040 (43.31)
-0.028 (-2.27) 69.4 0.805 5.799 0.604 0.454 0.158
III 0.331 (25.08)
-0.342 (-1.44) 8.3 0.699 5.048 0.608 0.306 0.184
Model R2 = 0.674633 F-test 4.91** 47.07** 3.18* 41.78** 6.44**
Panel C: Accruals+ = α + βTOTFEE
Cluster α β % Sample Book-to-Market
Log (Market Value)
Board Composition
Institutional Holdings
Insider Holdings
I 0.131 (26.97)
-0.114 (-2.47) 24.5 0.709 5.399 0.593 0.415 0.183
II 0.043 (36.37)
-0.024 (-1.62) 72.2 0.797 5.889 0.599 0.469 0.156
III 0.354 (11.71)
-0.258 (-0.43) 3.3 0.616 5.282 0.559 0.312 0.174
Model R2 = 0.619948 F-test 3.73* 16.44** 1.02 15.48** 2.99
Panel D: Accruals- = α + βTOTFEE
Cluster α β % Sample Book-to-Market
Log (Market Value)
Board Composition
Institutional Holdings
Insider Holdings
I -0.144 (-20.85)
0.286 (2.66) 18.9 0.758 4.862 0.574 0.349 0.188
II -0.326 (-21.96)
0.349 (1.30) 14.8 0.722 4.997 0.617 0.310 0.196
III -0.036 (-24.56)
0.030 (1.36) 66.3 0.819 5.688 0.611 0.434 0.159
Model R2 = 0.707190 F-test 1.95 32.76** 3.73* 24.94** 3.49*
** (*) Significant at the 1% (5%) level. Model R2 is computed as 1 – L/L0 , where L is the maximum for the log-likelihood for the three cluster solution and L0 is log-likelihood for the null model (intercept only).
RATIO is the ratio of fees paid to auditors for non-audit services divided by the total fees (the sum of audit and non-audit fees) paid to auditors. TOTFEE is the ratio of total fees (the sum of audit and non-audit fees) paid to the auditor, to the total revenue received that year by the auditor. NONAUDFEE is the ratio of non-audit fees paid to the auditor, to the total revenue received that year by the auditor. Book-to-Market (BM) is the book to market ratio calculated as the book value of common equity (item 60) divided by market capitalization at the end of the fiscal year (item 25 * item 199). Board Composition is the fraction of directors serving on the board who are independent from management. Institutional Holdings is the fraction of outstanding shares that are held by institutions (as reported by WorldScope). Insider Holdings is the fraction of outstanding shares that are held by insiders (as reported by WorldScope). The accrual model is estimated using the Jones (1991) technique of decomposing total accruals into a normal (expected) and abnormal (unexpected) component. The method of decomposition is as follows: TA = α + β1(∆Sales-∆REC) + β2PPE + β3BM + β4CFO + ε (1) TA is the difference between operating cash flows (item 308) and income before extraordinary items (item 123) as reported on the statement of cash flows. ∆Sales is the change in sales (item 12) for the year. ∆REC is the change in receivables reported on the statement of cash flows (item 302) for the year. PPE is the gross amount of property, plant and equipment (item 7). CFO is the operating cash flows (item 308). All variables used in the abnormal accrual model (except BM) are scaled by average total assets using assets from the start and end of the fiscal year. Accruals is the residual from equation (1) above, |Accruals| is the absolute value of the residual from equation (1) above. Accruals+ is equal to Accruals when Accruals>0 and zero otherwise. Accruals- is equal to Accruals when Accruals<0 and zero otherwise.
Table 8
Latent Class Mixture Analysis – Using NONAUDFEE.
Panel A: Accruals = α + βNONAUDFEE
Cluster α β % Sample Book-to-Market
Log (Market Value)
Board Composition
Institutional Holdings
Insider Holdings
I 0.022 (7.19)
-0.004 (-0.06) 45 0.771 5.476 0.588 0.428 0.175
II 0.016 (8.43)
-0.021 (-0.57) 36 0.810 5.923 0.609 0.462 0.153
III -0.084 (-7.95)
0.396 (1.05) 19 0.711 4.949 0.607 0.322 0.187
Model R2 = 0.089009 F-test 3.51* 64.90** 4.72** 47.30** 7.31**
Panel B: |Accruals| = α + βNONAUDFEE
Cluster α β % Sample Book-to-Market
Log (Market Value)
Board Composition
Institutional Holdings
Insider Holdings
I 0.133 (33.06)
-0.206 (-3.65) 22.2 0.726 5.231 0.582 0.397 0.187
II 0.039 (44.06)
-0.038 (-2.25) 69.4 0.806 5.797 0.605 0.454 0.158
III 0.330 (25.68)
-0.529 (-1.46) 8.4 0.702 5.033 0.608 0.304 0.185
Model R2 = 0.674277 F-test 4.99** 46.07** 3.42* 42.59** 6.52**
Panel C: Accruals+ = α + βNONAUDFEE
Cluster α β % Sample Book-to-Market
Log (Market Value)
Board Composition
Institutional Holdings
Insider Holdings
I 0.130 (26.88)
-0.151 (-2.25) 24.5 0.710 5.392 0.591 0.415 0.181
II 0.042 (36.96)
-0.033 (-1.60) 72.2 0.797 5.890 0.599 0.469 0.156
III 0.357 (11.61)
-0.646 (-0.46) 3.3 0.599 5.298 0.568 0.314 0.176
Model R2 = 0.620519 F-test 3.91* 16.68** 0.75 15.17** 2.66
Panel D: Accruals- = α + βNONAUDFEE
Cluster α β % Sample Book-to-Market
Log (Market Value)
Board Composition
Institutional Holdings
Insider Holdings
I -0.142 (-20.42)
0.334 (2.65) 18.9 0.754 4.879 0.574 0.348 0.189
II -0.325 (-22.41)
0.470 (1.26) 14.8 0.726 4.982 0.617 0.312 0.195
III -0.036 (-25.05)
0.043 (1.40) 66.3 0.819 5.687 0.611 0.434 0.160
Model R2 = 0.706307 F-test 1.94 32.12** 3.82* 25.16** 3.76*
** (*) Significant at the 1% (5%) level. Model R2 is computed as 1 – L/L0 , where L is the maximum for the log-likelihood for the three cluster solution and L0 is log-likelihood for the null model (intercept only).
RATIO is the ratio of fees paid to auditors for non-audit services divided by the total fees (the sum of audit and non-audit fees) paid to auditors. TOTFEE is the ratio of total fees (the sum of audit and non-audit fees) paid to the auditor, to the total revenue received that year by the auditor. NONAUDFEE is the ratio of non-audit fees paid to the auditor, to the total revenue received that year by the auditor. Book-to-Market (BM) is the book to market ratio calculated as the book value of common equity (item 60) divided by market capitalization at the end of the fiscal year (item 25 * item 199). Board Composition is the fraction of directors serving on the board who are independent from management. Institutional Holdings is the fraction of outstanding shares that are held by institutions (as reported by WorldScope). Insider Holdings is the fraction of outstanding shares that are held by insiders (as reported by WorldScope). The accrual model is estimated using the Jones (1991) technique of decomposing total accruals into a normal (expected) and abnormal (unexpected) component. The method of decomposition is as follows: TA = α + β1(∆Sales-∆REC) + β2PPE + β3BM + β4CFO + ε (1) TA is the difference between operating cash flows (item 308) and income before extraordinary items (item 123) as reported on the statement of cash flows. ∆Sales is the change in sales (item 12) for the year. ∆REC is the change in receivables reported on the statement of cash flows (item 302) for the year. PPE is the gross amount of property, plant and equipment (item 7). CFO is the operating cash flows (item 308). All variables used in the abnormal accrual model (except BM) are scaled by average total assets using assets from the start and end of the fiscal year. Accruals is the residual from equation (1) above, |Accruals| is the absolute value of the residual from equation (1) above. Accruals+ is equal to Accruals when Accruals>0 and zero otherwise. Accruals- is equal to Accruals when Accruals<0 and zero otherwise.
Table 9
Audit fee regression models for our sample of firm observations with available Compustat and Standard and Poors audit fee data in 2000 and 2001. Regressions
are run separately for each industry group.
Log(Fee) = φ0 + φ1Log(Assets) + φ2Log(Segments) + φ3Inventory + φ4Receivables + φ5Debt + φ6Income + φ7LOSS + φ8Opinion + ε (4)
Panel A: Total Fees
Independent Variable
Mean Estimate
T Statistic (minimum)
T Statistic (maximum)
Intercept 2.154 6.19 45.49 Log(Assets) 0.626 13.15 96.63
Log(Segments) 0.156 0.97 7.13 Inventory 1.187 1.05 7.72
Receivables 1.250 0.04 0.29 Debt 0.120 0.19 1.40
Income -0.286 -0.67 -4.92 LOSS 0.155 0.72 5.29
Opinion 0.187 1.02 7.50
Mean Adjusted R2 0.749
Panel B: Non-audit Fees
Independent Variable
Mean Estimate
T Statistic (minimum)
T Statistic (maximum)
Intercept 0.042 0.38 2.79 Log(Assets) 0.835 9.13 67.09
Log(Segments) 0.155 0.31 2.28 Inventory 0.837 0.52 3.82
Receivables -0.537 -0.25 -1.84 Debt -0.002 0.03 0.22
Income 0.143 -0.20 -1.47 LOSS 0.188 0.41 3.01
Opinion 0.072 0.28 2.06
Mean Adjusted R2 0.587
Mean coefficients are based on industry level regressions, reported T-statistics (minimum) are the mean t-statistics across the industry level regressions, and T-statistics (maximum) are the mean t-statistics across the industry level regressions multiplied by the square root of the number of industries used to compute the mean ( 54 ). The estimated residual from equation (4) is our proxy measure for “abnormal” fees. To transform this to a dollar amount we raise exp to the power of the predicted value of Log(Fee) and then subtract this value from the dollar fee. The result (after deflating by auditor firm revenue) is denoted as ABTOTFEE (ABNONAUDFEE) for total fees and non-audit fees.
Log(Assets) is the natural logarithm of total assets (Compustat data item # 6). Log(Segments) is the natural logarithm of the number of business segments reported on the Compustat Segment Data File. Inventory is the ratio of the dollar value of inventory (item 3) to total assets (item 6). Receivables is the ratio of the dollar value of accounts receivable (item 2) to total assets (item 6). Debt is the sum of short term debt (item 34) and long term debt (item 9) to total assets (item 6). Income is the ratio of operating income after depreciation (item 178) to average total assets (item 6). LOSS is an indicator variable equal to one if the firm reports negative Income in any of the previous three years and zero otherwise. Opinion is an indicator variable equal to one of the firm receives a qualified audit opinion and zero otherwise. A qualified audit opinion is defined as anything other than the standard unqualified audit opinion coded as “one” by Compustat.
Table 10
Latent Class Mixture Analysis - Abnormal audit fees.
Panel A: Accruals = α + β1HighABTOTFEE + β2LowABTOTFEE
Cluster α β1 β2 %
Sample
I -0.084 (-7.63)
0.584 (0.82)
-0.946 (-0.94) 19.1
II 0.021 (7.23)
0.031 (0.29)
-0.134 (-0.37) 48.2
III 0.016 (7.57)
0.014 (0.23)
0.186 (0.91) 32.7
Model R2 = 0.086791 Panel B: |Accruals| = α + β1HighABTOTFEE + β2LowABTOTFEE
Cluster α β1 β2 %
Sample
I 0.136 (73.48)
-0.228 (-4.20)
1.301 (6.63) 22.1
II 0.040 (65.45)
-0.044 (-2.43)
0.184 (3.06) 69.6
III 0.331 (39.33)
-0.689 (-2.02)
0.413 (0.47) 8.3
Model R2 = 0.676026 Panel C: Accruals- = α + β1HighABTOTFEE + β2LowABTOTFEE
Cluster α β1 β2 %
Sample
I -0.147 (-21.47)
0.398 (2.08)
-1.854 (-3.12) 19.9
II -0.036 (-25.03)
0.069 (1.30)
-0.210 (-1.40) 66.4
III -0.331 (-20.92)
0.558 (0.74)
0.739 (0.50) 13.7
Model R2 = 0.717128 Panel D: Accruals+ = α + β1HighABTOTFEE + β2LowABTOTFEE
Cluster α β1 β2 %
Sample
I 0.131 (55.06)
-0.175 (-2.75)
0.721 (2.80) 24.2
II 0.043 (53.78)
-0.032 (-1.50)
0.154 (1.90) 72.4
III 0.359 (17.57)
-0.691 (-0.86)
2.048 (1.02) 3.4
Model R2 = 0.622091
Panel E: Accruals = α + β1HighABNONAFEE + β2LowABNONAFEE
Cluster α β1 β2 %
Sample
I 0.017 (16.46)
-0.026 (-0.66)
0.325 (3.24) 32.9
II -0.084 (-9.66)
0.799 (1.71)
-1.167 (-1.04) 19.1
III 0.022 (11.39)
0.015 (0.18)
0.026 (0.12) 48.0
Model R2 = 0.087216 Panel F: |Accruals| = α + β1HighABNONAFEE + β2LowABNONAFEE
Cluster α β1 β2 %
Sample
I 0.135 (33.18)
-0.299 (-2.29)
1.453 (4.00) 22.1
II 0.040 (43.80)
-0.086 (-2.36)
0.242 (2.32) 69.6
III 0.330 (24.96)
-0.950 (-1.08)
0.474 (0.40) 8.3
Model R2 = 0.674661 Panel G: Accruals- = α + β1HighABNONAFEE + β2LowABNONAFEE
Cluster α β1 β2 %
Sample
I -0.146 (-21.96)
0.562 (2.21)
-2.219 (-2.19) 19.2
II -0.036 (-25.11)
0.103 (1.45)
-0.210 (-1.20) 66.5
III -0.328 (-22.28)
0.877 (1.00)
0.361 (0.29) 14.3
Model R2 = 0.712791 Panel H: Accruals+ = α + β1HighABNONAFEE + β2LowABNONAFEE
Cluster α β1 β2 %
Sample
I 0.043 (37.24)
-0.080 (-1.89)
0.254 (1.90) 72.5
II 0.131 (26.58)
-0.227 (-1.45)
0.800 (2.00) 24.2
III 0.360 (11.14)
-0.945 (-0.31)
1.703 (0.32) 3.3
Model R2 = 0.621562
Model R2 is computed as 1 – L/L0 , where L is the maximum for the log-likelihood for the three cluster solution and L0 is log-likelihood for the null model (intercept only). Log(Fee) = φ0 + φ1Log(Assets) + φ2Log(Segments) + φ3Inventory + φ4Receivables + φ5Debt + φ6Income + φ7LOSS + φ8Opinion + ε (4) The estimated residual ( ε̂ ) from the above equation is our proxy measure for “abnormal” fees. To transform this to a dollar amount we raise exp to the power of the predicted value of Log(Fee) and then subtract this value from the dollar fee. The result (after deflating by auditor firm revenue) is denoted as ABTOTFEE (ABNONAUDFEE) for total fees and non-audit fees. Log(Assets) is the natural logarithm of total assets (Compustat data item # 6). Log(Segments) is the natural logarithm of the number of business segments reported on the Compustat Segment Data File. Inventory is the ratio of the dollar value of inventory (item 3) to total assets (item 6). Receivables is the ratio of the dollar value of accounts receivable (item 2) to total assets (item 6). Debt is the sum of short term debt (item 34) and long term debt (item 9) to total assets (item 6). Income is the ratio of operating income after depreciation (item 178) to average total assets (item 6). LOSS is an indicator variable equal to one if the firm reports negative Income in any of the previous three years and zero otherwise. Opinion is an indicator variable equal to one of the firm receives a qualified audit opinion and zero otherwise. A qualified audit opinion is defined as anything other than the standard unqualified audit opinion coded as “1” by Compustat. HighABTOTFEE (HighABNONAUDFEE) is equal to ABTOTFEE (ABNONAUDFEE) when ABTOTFEE (ABNONAUDFEE ) is greater than zero, and zero otherwise. LowABTOTFEE (LowABNONAUDFEE) is equal to ABTOTFEE (ABNONAUDFEE) when ABTOTFEE (ABNONAUDFEE ) is less than or equal to zero, and zero otherwise. Book-to-Market (BM) is the book to market ratio calculated as the book value of common equity (item 60) divided by market capitalization at the end of the fiscal year (item 25 * item 199). Board Composition is the fraction of directors serving on the board who are independent from management. Institutional Holdings is the fraction of outstanding shares that are held by institutions (as reported by WorldScope). Insider Holdings is the fraction of outstanding shares that are held by insiders (as reported by WorldScope). The accrual model is estimated using the Jones (1991) technique of decomposing total accruals into a normal (expected) and abnormal (unexpected) component. The method of decomposition is as follows: TA = α + β1(∆Sales-∆REC) + β2PPE + β3BM + β4CFO + ε (2) TA is the difference between operating cash flows (item 308) and income before extraordinary items (item 123) as reported on the statement of cash flows. ∆Sales is the change in sales (item 12) for the year. ∆REC is the change in receivables reported on the statement of cash flows (item 302) for the year. PPE is the gross amount of property, plant and equipment (item 7). CFO is the operating cash flows (item 308). All variables used in the abnormal accrual model (except BM) are scaled by average total assets using assets from the start and end of the fiscal year.